🚀 Azure DevOps MCP Server

A custom Model Context Protocol (MCP) server built with NestJS, enabling AI agents (like Claude Desktop) to securely query Azure DevOps data such as work items, projects, and ticket statistics.

📌 Overview

This project exposes Azure DevOps data through a standards-based MCP server, allowing any MCP-compatible AI agent to perform real-time queries such as:

• 🔍 List all projects in the organization
• 📝 Get count of created / active / closed work items
• 🧩 Fetch ticket details
• 📊 Query backlog items or sprints
• 🏗️ Extend Azure DevOps automation through AI

It supports two modes:

1️⃣ STDIO MCP Server

Tested locally using the MCP Inspector (not using a direct Claude Desktop STDIO client).

2️⃣ HTTP MCP Server

Allows remote access using MCP-over-HTTP.

🏗️ Architecture

NestJS Application
│
├── MCP Module
│   ├── Tool Definitions
│   ├── STDIO Transport
│   └── HTTP Transport
│
├── Azure DevOps Service
│   └── Work Item API
│   └── Project API
│
└── Utilities
    ├── MCP Message Handlers
    └── Logger

✨ Features

✔️ Custom Tools for Azure DevOps

Tools exposed to MCP clients:

✔️ STDIO Transport (MCP Inspector)

This project was tested locally using the MCP Inspector, which acts as an MCP client that connects over STDIO to your server.

Run the server so the inspector can connect over STDIO:

node dist/main.js

Start the inspector with:

npx @modelcontextprotocol/inspector

When prompted in the inspector, choose the STDIO connection and point it at the running process (the inspector will spawn or attach to the process as configured).

✔️ HTTP MCP Support

Start server:

npm run start:prod

You can POST MCP-style requests:

POST http://localhost:3000/mcp

🔧 Installation

git clone https://github.com/saktheeswar/Azure_Devops_MCP_Server.git
cd azure-devops-mcp
npm install
npm run build

🔑 Environment Variables

Create .env:

BASE_URL=url
AUSER_NAME=your-personal-access-token
ADO_PAT=your-default-project

▶️ Running the Server

Run in STDIO mode (for MCP Inspector)

node dist/main.js

Run in HTTP mode

npm run start:prod

🧪 Testing with MCP Inspector

(You will explain this with screenshots in your blog.)

npx @modelcontextprotocol/inspector

Then connect using STDIO:

node dist/main.js

💻 Code Walkthrough

📁 MCP Server Setup

const server = new Server({
  name: "azure-devops-mcp",
  version: "1.0.0",
  tools: {
    list_projects: { ... },
    count_tickets: { ... }
  }
});

📁 Azure DevOps API Integration

async getProjects() {
  const url = `${this.baseUrl}/projects?api-version=7.0`;
  return this.http.get(url, this.headers);
}

📁 STDIO Bootstrap

bootstrapStdio(server);

📁 HTTP Controller

@Post('/mcp')
handleMcp(@Body() body) {
  return this.mcpHttpService.process(body);
}

📚 Example MCP Tool Call

{
  "method": "tools/list_projects",
  "params": {}
}

Response:

{
  "projects": [
    { "name": "Frontend" },
    { "name": "Backend" },
    { "name": "Infrastructure" }
  ]
}

🧩 Use Cases

For Developers

• Query Azure DevOps without opening the UI
• Get ticket summaries instantly
• Automate repetitive DevOps activities

For AI Assistants

• Smart sprint planning
• Ticket prioritization
• Automated status reporting

For Teams

• Faster decision making
• AI-powered insights

🔥 Future Enhancements

• Create/Update work items using AI
• Integration with Release Pipelines
• Sprint burndown insights
• PR and Repository analytics

📎 Screenshots (Add your own)

• Claude Desktop working
• MCP Inspector connected
• Tool response logs

🧑‍💻 Author

Saktheeswaran M AI Engineer & Full-Stack Developer

🔗 Source Code

👉 GitHub Repository: (https://github.com/saktheeswar/Azure_Devops_MCP_Server)